fixed array in struct instead of heap slice

[brianolson]

No description provided.

[brianolson]

winning benchmarks, from go test -memprofile /tmp/cgbench.pprof -run xxx -bench BenchmarkMarshaling

branch
BenchmarkMarshaling-16    	 1705956	       732.9 ns/op	      48 B/op	       1 allocs/op
2741714 48B objects, 1.6 alloc/op

master
BenchmarkMarshaling-16    	 1331407	       863.6 ns/op	      80 B/op	       2 allocs/op
4898996 objects, 3.68 alloc/op (2588711 16B + 2310285 64B)

[brianolson]

Applying the same strategy to reading and unmarshalling works, but is smaller because the test objects are lots of allocations. NewCborReader() setup gets better, but is swamped by the rest of of the object unmarshal, but is still an easy improvement.

go test -memprofile /tmp/cgbench.pprof -run xxx -bench BenchmarkUnmarshaling -v -gcflags -m 2> /tmp/go_build_debug

branch
BenchmarkUnmarshaling-16    	  207495	      6459 ns/op	    1772 B/op	      21 allocs/op
NewCborReader 229383 32B (1.11 alloc/op)

master
BenchmarkUnmarshaling-16    	  180823	      6467 ns/op	    1796 B/op	      22 allocs/op
NewCborReader 152924 48B + 131074 16B (1.57 alloc/op)